Not really a bioinformatics question, I think.

I good thing about open source is that you can download the code and check it yourself. I did so and found the following code in pheatmap.r file:

scale_rows = function(x){
    m = apply(x, 1, mean, na.rm = T)
    s = apply(x, 1, sd, na.rm = T)
    return((x - m) / s)
}

scale_mat = function(mat, scale){
    if(!(scale %in% c("none", "row", "column"))){
        stop("scale argument shoud take values: 'none', 'row' or 'column'")
    }
    mat = switch(scale, none = mat, row = scale_rows(mat), column = t(scale_rows(t(mat))))
    return(mat)
}

The second function chooses an action depending on the value of the variable scale. The first function performs row-wise scaling (or col-wise on the transposed matrix). So, scale in this package means removing the mean (centering) and dividing by the standard deviation (scaling). Also, take a look at ?scale for the base function. Why the author didn't use this function is not clear to me.